Why does my Firestore vector search return a match for an unrelated query 'foo'?

I'm using Firestore to store and retrieve vector embeddings for text queries, but I'm getting unexpected matches. I've embedded the phrase "What's the best bet on the table" using GoogleAI's textEmbeddingGecko001 and saved it to Firestore. My goal is to match similar sentences (e.g., "What's the TOP bet on the table") while unrelated queries like "foo" should not match. However, when I query with "foo", I still get a matching document.

Below is the code used for saving the document:

JAVASCRIPTCopied
require('dotenv').config({ path: __dirname + '/.env' })

const { initializeApp, cert } = require('firebase-admin/app');
const { getFirestore, FieldValue } = require('firebase-admin/firestore');
const serviceAccount = require('./secrets/crapsai-72b89-firebase-adminsdk-pnqva-e321c8eecb.json');

const { gemini15Flash, googleAI } = require('@genkit-ai/googleai');
const { textEmbeddingGecko001 } = require('@genkit-ai/googleai');

const { genkit } = require('genkit');

const ai = genkit({
    plugins: [
        googleAI({
            apiKey: process.env.GOOGLE_GENAI_API_KEY
        })
    ],
    model: gemini15Flash
});

const app = initializeApp({
    credential: cert(serviceAccount)
});

const firestore = getFirestore(app);

var question = "What's the best bet on the table";

(async () => {
    try {

        const embedding = await ai.embed({
            embedder: textEmbeddingGecko001,
            content: question
        });

        console.log("Got the embeddings");
        console.log(embedding);

        await firestore.collection("questions").add({
            "question": question,
            "embedding": FieldValue.vector(embedding),
            "confirmed": true
        });

        console.log("DONE");
        process.exit(0);
    } catch (e) {
        console.log("Error", e);
        process.exit(1);
    };
})();

Below is the retrieval code:

JAVASCRIPTCopied
require('dotenv').config({ path: __dirname + '/.env' })

const { initializeApp, cert } = require('firebase-admin/app');
const { getFirestore } = require('firebase-admin/firestore');
const serviceAccount = require('./secrets/crapsai-72b89-firebase-adminsdk-pnqva-e321c8eecb.json');

const { gemini15Flash, googleAI, textEmbeddingGecko001 } = require('@genkit-ai/googleai');

const { defineFirestoreRetriever } = require('@genkit-ai/firebase');

const { genkit } = require('genkit');

const ai = genkit({
    plugins: [
        googleAI({
            apiKey: process.env.GOOGLE_GENAI_API_KEY
        })
    ],
    model: gemini15Flash
});

const app = initializeApp({
    credential: cert(serviceAccount)
});

const firestore = getFirestore(app);

const retriever = defineFirestoreRetriever(ai, {
    name: 'questionRetriever',
    firestore,
    collection: 'questions',
    contentField: 'question',
    vectorField: 'embedding',
    embedder: textEmbeddingGecko001,
    distanceMeasure: 'COSINE'
});

var question = "foo";

(async () => {
    try {
        const docs = await ai.retrieve({
            retriever,
            query: question,
            options: {
                limit: 5,
                where: { confirmed: true }
            },
        });

        docs.forEach(doc => {
            console.log(doc);
        })

        console.log("DONE");
        process.exit(0);
    } catch (e) {
        console.log("Error", e);
        process.exit(1);
    };
})();

I expected that querying with "foo" would return no matches, but it still returns a document. Is there an issue with my setup or am I misunderstanding how vector search works in Firestore?

Additional context: Firestore's vector search returns the top K nearest neighbors by default, even if they aren't sufficiently similar. It appears I might need to implement a similarity threshold in my query to filter out false matches. Any guidance on how to approach this would be appreciated.